Our investigation of how to build low noise power supplies for custom electronic boards
Modern electronic circuits often require several low voltage DC power supplies. It is normally convenient to provide the board with only one DC input voltage, often 5V or 12V and to generate the other power supply levels from that. For a system that has an AC mains input, it is usual to buy in, or use a standard design to generate a single DC voltage from that. That isolates the mains voltage problems and regulations from your low voltage board design. After that you have the same issues for your multiple lower voltage circuits.
The design of these power supply circuits is an important part of almost any custom electronics design, but is often treated as something to be thrown together without much effort or care. Actually getting the design of the power supplies correct can win a lot of performance and reliability for the final product, and can save lots of difficult hours spent debugging problems that can be caused by bad Power Supply design.
Because of the importance that we assign to this part of design, we made a thorough study of PSU design in the particular context of the board level designs that we are normally involved in. For our applications we wanted to have a particular focus on low noise and ripple, in as small a footprint as possible - being less concerned about the last few % of efficiency.
What type of power circuit should I choose?
The simplest way to generate a supply that is lower in voltage than the voltage of the incoming DC voltage is a linear regulator. The linear regulator device is simply a voltage divider that maintains its output voltage over a range of currents demanded of it. With the correct use of bypass and bulk storage capacitors it is quite simple to get the voltage that you need, and to have low noise. The problem with this voltage divider is that the same current is drawn from the input supply as for the load. This means that the power defined as [current x (input_voltage – load_voltage)] must be dissipated (mainly as hgeat) by the regulator device. For high powers this means the regulator will become hot, and in order to dissipate high power the devices must be manufactured in a large package.
The result of this is that we choose to only use linear regulators in situations with very low current requirements – maybe just a few mA. Then we can select small footprint devices that will not need to become too hot.
The solution for this power dissipation problem is to use a Switch Mode Power Supply (SMPS) circuit. Here the input DC voltages is switched on and off across an inductor or capacitor. During the on period, energy is stored in the inductor or capacitor. During the off period, that energy is released to the output in a controlled manner. By changing the ratio of the on period and off period, the output voltage can be controlled. Because the input voltage is switched on and off, the input current is not constant. It can be averaged using input capacitors as bulk energy storage, the resulting level of input current is not forced to be the same as the load current. Using this technique results in a Power Supply circuit that can be in the region of 95% efficient.
Switch mode supplies can also be used to provide step down (Buck), step up (Boost) and negative (Inverting) voltages. This makes them a very valuable tool to have in our arsenal.
Switch Mode Power Supplies - how to get it right
As Engineers, we had in the past been forced by time pressures to simply follow design guidelines on data sheets and application notes. The results that we achieved were distinctly variable and mostly unsatisfying, so because we are engineers, we wanted to understand the issues thoroughly. So we invested a significant time to study and test our own theories regarding their designs.
Switched capacitor designs are useful for low supply currents, where we might be able to use a linear regulator. We are normally required to provide several amps (sometimes 10s of amps) so we chose to focus our studies on switched Inductor designs. We wanted the results of our investigation to give us guidelines of how to design real life products, so we wanted to work with commercially available SMPS control chips.
In order to make our study, we needed detailed data and design details of the controller devices. Having studied the information available from several suppliers of such controllers, we chose to use devices from Linear Technology Corporation because of their wide range of devices and excellent data sheets.
The Buck, Boost and Inverting supplies do have differences, but have a lot of similarities. Our most used supply circuit is a Buck, so while most of the rest of this page will be apply to all types, it will be from the perspective of the Buck converter.
The following diagram shows the key components of a Buck Converter.
For a typical Buck circuit, two FETs are required. The type of FET required is an N-channel enhancement mode MOSFET.
In the diagram, the top FET is switched on during the ON period and switched off during the OFF period. This FET is referred to as either the ‘top’ FET or the ‘control’ FET.
In the diagram, the bottom FET is switched off during the ON period and switched on during the OFF period. This FET is referred to as either the ‘bottom’ FET or the ‘sync’ FET.
During the ON time, the top FET switch is closed, connecting the left hand side of the inductor to Vin. The voltage across the inductor, assuming no voltage drop across the FET, becomes Vin – Vout. During this time, the bottom FET is open and no current flows through the bottom FET branch of the circuit.
The inductor current IL, which is positive, increases linearly according to the equation:
VL = L x di/dt
During the OFF time, the top FET is switched open and the bottom FET is closed. The voltage across the inductor (assuming no FET voltage drop) becomes –Vout.
Consequently, during the off time, the decrease in inductor current is:
di = ((-Vout) x dt) / L
The converter is said to be operating in continuous mode if the current through the inductor never falls to zero during the commutation cycle (that is, during the switching period T).
A buck controller maintains the required output voltage by controlling the duty cycle while it is switching in continuous mode. As the duty cycle increases, Vout increases for the same input voltage. The control loop of the converter will monitor the output voltage. As the load increases, the output voltage will drop. To compensate for the drop in supply the controller will increase the duty cycle by a small margin to result in a net increase in the DC current.
The inductor current is a triangular waveform with a DC component and an AC ripple component, as depicted in the diagram below.
The DC level of the inductor is the DC output current supplied to the load. In an ideal power supply circuit, the output current is a pure DC waveform. For this to be achieved, the output capacitor must filter the entire AC component of the inductor current.
When designing a supply it is desirable to maintain continuous mode for all of the expected levels of DC output current. In discontinuous mode, the converter performs a small number of switching cycles and then switches into an idle state with all switches open for a long period of time. This introduces switching waveforms that have an unpredictable and widely varying frequency content – really an enemy of our aims to control the noise and ripple associated with this part of our design.
For continuous mode to be used at a particular load level, the converter must be built around an inductor that guarantees the ripple current does not reach zero at that load.
When designing a switching power supply a control function is required that will drive the on time and off time of the supply circuit. This control function must accurately monitor the input and output voltages in order to maintain a stable supply at a steady output level despite variations in input voltage and variations in load.
The best solution for the control function will always be a dedicated IC. There are a very large number of power supply ICs that are available for switching power supply design. It is very important however to choose the right device as there as many badly designed control ICs as there are good designs. We consider that the devices from Linear Technology are some of the best we can find.
Power supply control ICs fall into two categories, Controllers and Converters. A controller is an IC that has the control loop circuitry necessary to control the switching of the supply. A converter is an IC that has control circuitry and in addition provides the power FETs necessary for switching.
Converter ICs offer a simpler solution with less external components. However, as the power FETs are built into the same IC as the control function there is usually a current limit in the order of just a few amps. The result is that for supplies with high current demands a controller will always be necessary.
Controllers and converters are available using either Pulse Width Modulation (PWM) control or Pulse Frequency Modulation (PFM) control.
The PWM scheme is the scheme that we have been discussing. This scheme operates with a continuously running switching function at a specific frequency where the pulse width (or duty cycle), is modified to compensate for variations in load.
The PFM scheme operates by changing the frequency of pulses in order to compensate for variations in load.
The advantage of PFM is low quiescent current, which means good efficiency for small loads. Unfortunately, the downside is more difficult component choice and much higher noise.
With PWM, the ripple is usually smaller and therefore easier to filter by the output capacitor. The higher the switching frequency, the smaller the inductor and capacitor components can become. In addition, the interference created by a PWM based supply is limited to the switching frequency and harmonics of that frequency. With a PFM approach however, there can be interference over a much wider frequency range.
Whether a PFM scheme or PWM scheme is best will depend on the requirements of the load. For devices that are performing analogue processing such as A-to-D converters, it is important to keep the switching frequency out of the signal frequency band where possible. The higher the switching frequency, the easier it is to build an effective filter.
All switched mode power supplies use a closed loop control circuit to maintain the output voltage at the desired level as the power supply load changes. In simple terms, the control loop must detect load changes and supply more current when the load increases and less when the load decreases.
All switched mode converter and controller ICs employ an op-amp circuit to monitor and respond to load changes reflected by the feedback voltage of the supply. The amplifier used in these circuits is commonly referred to as the ‘Error Amplifier’. Designing a properly working amplifier-based closed loop network requires a proper understanding of the issues of gain and phase.
There are two main elements in creating a design that works well in response to a step change in load.
The first principle is to create a circuit that causes the minimum deviation in output voltage for a step change in load, while recovering to the steady state condition as quickly and smoothly as possible. This principle is a principle of quality of response.
The second principle is to respond to any step change without entering oscillation of the control loop. This principle is a principle of basic loop stability and will be far more serious to correct power supply operation if not properly observed.
In general, all switched mode control loops are built to operate as low pass filters, with respect to the feedback voltage. That is, from DC up to frequencies less than the switching frequency of the supply, changes in the load must be amplified by the control loop, resulting in the necessary change in output current. Beyond this frequency band, the feedback signal must be attenuated to prevent the control loop incorrectly modifying current due to high frequency switching noise in the circuit.
The ideal frequency response has a flat gain level from DC to the corner frequency. This level is referred to as the DC Gain of the closed loop. At the corner frequency there is a pole that begins a 20db/decade drop in gain as the frequency increases.
In the ideal circuit a single pole exists, causing a simple –20db/decade gain drop. In practice, a control loop will contain several poles and zeros due to the internal silicon design of the controller or converter IC, as well as the output capacitor, output capacitor ESR and load resistance. All of these extra poles and zeros create a far different response from the ideal case.
The principle of Loop Compensation is to deliberately add additional poles and zeros to compensate for the behaviour of the unwanted poles and zeros that exist in the control loop, to bring the response back towards the ideal. This is usually done by the addition of a resistor and capacitor network to the output of the error amplifier. In some cases, capacitors may also be added to the voltage feedback network.
Choosing real components
We decided that we would standardise our designs around a few chosen controllers. For our most commonly used PSU design we would use an LTC3835 controller as it meets all of the needs we have identified in our study, and the excellent data allows us to make an intelligent spreadsheet tool that helps us to select real world components to go with it and comply with the sometimes conflicting requirements. This tool allows us to achieve good performance from our supply designs every time without having to remember all of the applicable equations intimately.
The type and size of inductor is typically the most important choice in a switched mode design. The size of inductor will dictate the frequency that can be used for switching and hence the frequencies at which noise will exist. It will also dictate the maximum current capability of the circuit and the required size of all other external components.
In general, when choosing an inductor it is important to select an inductance that is as big as possible. This will reduce ripple current, lower the peak to average ratio, lower the peak current, allow the converter to switch in continuous mode for a wider load range, and help to reduce output supply noise ripple levels. Unfortunately, as the inductance increases, the inductor package will physically increase in size and the current capability will drop, affecting the maximum output current that is possible. An inductor must therefore be chosen that balances ripple size with current rating to achieve the best balance of all elements.
Having as high a switching frequency as possible is normally an advantage for your design as the ripple and switching noise is correspondingly higher, and hence easier to remove while maintaining good circuit response at close to DC – allowing the circuit to react the to the varying demands of your board.
The LTC3835 controller also has the nice feature of being able to synchronise multiple supply circuits to the same clock, and to select the switching to occur at different phases of that clock. This means we can ensure that we do not have more than one PSU circuit switching at the same time. The peak noise from the different supplies will therefore not be added together, but rather distributed in time.
Converters all require adequate input and output capacitance to prevent voltage ripple appearing on the supplies and to prevent current transients causing spikes on the input supply.
The output capacitor of a switched mode converter has an important function to perform in reducing output ripple.
Using the buck converter example above, we can see that the current passing through the inductor is a triangle wave with a DC offset.
In an ideal power supply, the output current Iout is a constant DC level. Given the current function in the inductor, this creates the requirement that only the ILDC DC component of inductor current be passed to the output. That is, the capacitor must filter the entire AC component, returning this component via Ground back to the supply circuit input capacitance and FET switches.
For the entire AC component to be filtered by the output capacitor this must mean that the impedance of the capacitor is zero for all frequencies contained in the triangle wave of the inductor. In practice, the output capacitor will always have an impedance that is greater than zero due to the series resistance and there will also be a small inductance inherent in the packaging and an additional inductance created by the PCB layout. This means we much choose our capacitors and layout carefully to minimise these effects.
Switched mode converters will inject noise onto the input supply as well as the output supply. The input current will be almost identical to the Inductor current, so like the output capacitor, the input capacitor must also provide a low impedance for the ripple current. Again capacitor choice and layout is important.
The FETs must also be selected carefully because they must handle large currents and ideally, create no loss in the circuit. In practice there will be two main forms of loss in the FET of a switching converter.
The first form of loss is conduction loss due to the ON resistance of the FET. The lower the on resistance, the lower this loss. The second form of loss is due to switching currents as the FET moves back and forth between fully on and fully off. This second form of loss is typically smaller than the first, to the ON resistance of the FET is key.
The final choice of FETs is extremely complex as many parameters come into play – not all of which are available from their data sheets. Some choices must be made here about whether to favour low noise, or absolute efficiency of the Power Supply. In our case noise is normally the largest concern, as the efficiency losses that we are considering are normally only a few % difference overall – i.e. maybe 95% rather than 97%.
The templates that we have developed take all of these things into account and helps to select the values to the compensation circuit too.
Having the perfect selection of components does not guarantee that you will end up with a circuit that performs well, both in terms of response to the needs of the load, and the noise that gets generated and passed onto the rest of your design. The PCB layout is also critical.
We designed and manufactured a Power Supply test board, that allowed us to fit different FET types and to use different PCB layouts.
Using this board we tested and measured the relative performances allowing us to confirm that our theoretical understandings could be related to a real circuit.
This confirmed to us that the critical parameters meant that it is an advantage to use different FET types for the top and bottom FETs.
It also confirmed that using a particular placement strategy for the components and using poured copper to connect them on one layer – using a single connection point to the main power and GND planes of the main PCB, were very important.
Since making the spreadsheet tool, we have successfully designed tens of Switch Mode Power Supplies, and with extensive testing and measurement have been extremely pleased with the results.
If you have a need to design a Switch Mode Power supply then let us know, or more likely you would like to contact us about a complete custom board design, and you can rest assured that we can easily make an effective and low noise Power supply circuit to use in that design.